 |
Question about image processing capability
James from United States [3 posts] |
12 year
|
Hello,
I'm using Arduino controller and MATLAB for my autonomous vehicle project at my lab. I'm using two webcams and using image processing functions in MATLAB to do pursuit and evasion task where the vehicle follows certain colored objects autonomously while avoiding obstacles with different colors.
Basically I'm having 2 challenges for vision.
1) the image processing speed
I'm using hand-written image processing functions that identify objects based on pixel RGB values. Basically I use few sets of RGB threshold values and the blob-find function compares each pixel of the image from the camera to the threshold RGB value we gathered from pictures. The issue is that it takes a lot of time to do this threshold check and I'm wondering if Robo-realm's color-tracking software is much faster than what I'm using.
2) I'm using 2 webcams and I'm getting z-distances by looking at the height of the object identified. When the object gets too close the height of the object actually decreases as it gets out of the webcam's screen. I'm wondering if there are some features in Robo-realm's software that can continuously track this object and remember the object's id even when the geometric information changes suddenly.
Thank you
|
|
|
|
Anonymous |
12 year
|
James,
1) While I'm not sure about Matlabs vision processing I would image that just doing what you are doing in another language would speed things up quite a bit. The algorithm is not complex what you describe so it should function at a high fps assuming the underlying language is fast enough. If you implement what you describe in a c++ plugin dll for RoboRealm I'm sure that will perform quite well.
2) This will depend on the type of geometric changes. If you have a couple images to share that illustrate this issue we may be able to help more. Same goes for #1, if you have images and threshold ranges we can mock up something quick for you to compare speed with Matlab.
STeven.
|
|
|
|
James from United States [3 posts] |
12 year
|
Hello Steven,
Thank you for your reply.
I'm aware that writing my functions in C++ will speed things up quite a bit.
But I'm looking to see if there are some advanced features in Robo-realm
compared to my basic image processing functions that I have now.
I'm interested in buying this software if it can solve 2 problems
1) faster speed
As I said, I know that simply writing the whole code in another language will speed up things considerably. But I'm curious to know if there are any optimized algorithms in your color-tracking software. For instance if you know the position of slowly-moving colored target, you don't need to analyze each and every pixel of the image you get from webcam; you can only look at portion of the screen to analyze the blob which is more efficient in terms of speed.
2) Intelligent object-tracking
I'm curious to know if you have any built-in algorithms that track certain objects based on their positions. For instance, if the object is slowly moving you know rough position of the object unless it moves really fast in the image. So last position of the pixel compared to the next position of the pixel will help maintaining object identification (object id). So even if the object starts going to the edge of the screen and starts disappearing, you will always have information on certain object with ID #1, #2, etc.
I'm not sure if I phrased my questions effectively but I'd greatly appreciate if you can answer some of what I've mentioned.
James
|
|
|
|
Anonymous |
12 year
|
James,
1) Yes, there are optimizations such as what you refer to in order to speed up the processing. The specifics on the technique will depend on the module being used.
2) Yes, have a look at the Blob_Tracking module which has this capability.
http://www.roborealm.com/help/Blob_Tracking.php
If you can provide a sample video/images and desired output we can reply back with a robofile (RR configuration) that would help you along. With the 30 day trial you should be able to determine if this is an improvement in terms of speed and tracking or not prior to purchase. Best way to answer your questions is to see it for yourself.
STeven.
|
|
This Thread